persistent-memory-7layer

What problem does it solve?

Agents lose context between interactions and lack a structured, reliable way to store short-term working state, episodic conversation history, semantic embeddings, procedural SOPs, long-term facts, and calibration metrics; this Skill provides a unified 7-layer memory orchestration to preserve and surface relevant context across sessions.

Core Features & Use Cases

• Seven distinct layers (L1–L7): working memory (Redis 1h), short-term (Redis 24h), episodic (Supabase), semantic (pgvector embeddings), procedural (protocols/SOPs), long-term facts (Supabase), and meta-memory for calibration and metrics.
• Automated consolidation: nightly L2->L3->L6 consolidation, summarization of episodes into facts, and embedding of new facts into semantic memory for future retrieval.
• Semantic search & tooling: cosine-similarity search over pgvector/Qdrant, HNSW indexing, health checks, watcher events, and protocol-driven task automation (n8n/webhooks).
• Production-ready schema and migrations: SQL schema, triggers, row-level security policies, and seed protocols for briefing, meeting prep, consolidation, and critical alerts.
• Use case: Run MEMU to keep conversational context and user preferences available to your assistant across days, enable meeting prep that aggregates related history, and automatically surface long-term facts during decision workflows.

Quick Start

Install and configure MEMU with Redis, Supabase (pgvector), and Qdrant, set environment variables for API keys, and start the TypeScript Express service to enable the 7-layer memory and nightly consolidation.